Re: Fingerprint as cryptokey

Mike Amling wrote:
Yes, but is a unique codeword strictly necessary? The fingerprint data may be near the boundary of two or more regions. A cooperating "sender" would send only a codeword that is far from the boundary of its containing region.

I think the problem Francois Grieu describes is at least
theoretically interesting. No per-instance data besides the
readings of the print; multiple readers want to derive the
same key.

Nevertheless, I like your approach:

However, its possible to partition an N-dimensional space so that no point is close to more than N+1 regions.

Ah... that's a good thing to know, and I did not. One
dimension is obvious; in two dimensions, hexagons work and
so do courses of bricks; in three, shifted courses of bricks
look good. Maybe shifted-courses work indefinitely? Is there
a proffered general construction?

The fingerprint can thus limit the generated "key" to a set whose cardinality is linear in the number of dimensions measured. Trying all keys in the set can't take much more time than describing the point in N-space.

So the only extra instance-specific data we need is a hash of
the key (something that won't give away the key but allows us
to test candidate keys). Then the geometry of the partitioning
makes best-first key search fast. Not bad.



--
--Bryan
.

Relevant Pages

  • Re: Fingerprint as cryptokey
    ... Mike Amling wrote: ... so do courses of bricks; in three, shifted courses of bricks ... one or more dimensions are discrete, ...
    (sci.crypt)
  • Re: UKIP, Metrification and Low Railway Bridges
    ... The bricks were to an imperial set of dimensions. ... Trying to make some modifications was a real PITA as modern bricks are metric and don't match. ...
    (uk.railway)
  • Re: Alignments of the Newport Tower
    ... That's because the 'module' system enetered architectural design about ... The dimensions of bricks are designed to fit modules. ...
    (sci.archaeology)
  • Re: Fingerprint as cryptokey
    ... in two dimensions, hexagons work and ... so do courses of bricks; in three, ... Euclidean space R**N, but I'm not sure what I claimed is still true if one or more dimensions are discrete, rather than continuous. ...
    (sci.crypt)
Quantcast